fs/ext3/super.c

   1 /*
   2  *  linux/fs/ext3/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/config.h>
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/fs.h>
  23 #include <linux/time.h>
  24 #include <linux/jbd.h>
  25 #include <linux/ext3_fs.h>
  26 #include <linux/ext3_jbd.h>
  27 #include <linux/slab.h>
  28 #include <linux/init.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/parser.h>
  31 #include <linux/smp_lock.h>
  32 #include <linux/buffer_head.h>
  33 #include <linux/vfs.h>
  34 #include <linux/random.h>
  35 #include <linux/mount.h>
  36 #include <linux/namei.h>
  37 #include <linux/quotaops.h>
  38 #include <asm/uaccess.h>
  39 #include "xattr.h"
  40 #include "acl.h"
  41
  42 static int ext3_load_journal(struct super_block *, struct ext3_super_block *);
  43 static int ext3_create_journal(struct super_block *, struct ext3_super_block *,
  44                                int);
  45 static void ext3_commit_super (struct super_block * sb,
  46                                struct ext3_super_block * es,
  47                                int sync);
  48 static void ext3_mark_recovery_complete(struct super_block * sb,
  49                                         struct ext3_super_block * es);
  50 static void ext3_clear_journal_err(struct super_block * sb,
  51                                    struct ext3_super_block * es);
  52 static int ext3_sync_fs(struct super_block *sb, int wait);
  53 static const char *ext3_decode_error(struct super_block * sb, int errno,
  54                                      char nbuf[16]);
  55 static int ext3_remount (struct super_block * sb, int * flags, char * data);
  56 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf);
  57 static void ext3_unlockfs(struct super_block *sb);
  58 static void ext3_write_super (struct super_block * sb);
  59 static void ext3_write_super_lockfs(struct super_block *sb);
  60
  61 /*
  62  * Wrappers for journal_start/end.
  63  *
  64  * The only special thing we need to do here is to make sure that all
  65  * journal_end calls result in the superblock being marked dirty, so
  66  * that sync() will call the filesystem's write_super callback if
  67  * appropriate.
  68  */
  69 handle_t *ext3_journal_start_sb(struct super_block *sb, int nblocks)
  70 {
  71         journal_t *journal;
  72
  73         if (sb->s_flags & MS_RDONLY)
  74                 return ERR_PTR(-EROFS);
  75
  76         /* Special case here: if the journal has aborted behind our
  77          * backs (eg. EIO in the commit thread), then we still need to
  78          * take the FS itself readonly cleanly. */
  79         journal = EXT3_SB(sb)->s_journal;
  80         if (is_journal_aborted(journal)) {
  81                 ext3_abort(sb, __FUNCTION__,
  82                            "Detected aborted journal");
  83                 return ERR_PTR(-EROFS);
  84         }
  85
  86         return journal_start(journal, nblocks);
  87 }
  88
  89 /*
  90  * The only special thing we need to do here is to make sure that all
  91  * journal_stop calls result in the superblock being marked dirty, so
  92  * that sync() will call the filesystem's write_super callback if
  93  * appropriate.
  94  */
  95 int __ext3_journal_stop(const char *where, handle_t *handle)
  96 {
  97         struct super_block *sb;
  98         int err;
  99         int rc;
 100
 101         sb = handle->h_transaction->t_journal->j_private;
 102         err = handle->h_err;
 103         rc = journal_stop(handle);
 104
 105         if (!err)
 106                 err = rc;
 107         if (err)
 108                 __ext3_std_error(sb, where, err);
 109         return err;
 110 }
 111
 112 void ext3_journal_abort_handle(const char *caller, const char *err_fn,
 113                 struct buffer_head *bh, handle_t *handle, int err)
 114 {
 115         char nbuf[16];
 116         const char *errstr = ext3_decode_error(NULL, err, nbuf);
 117
 118         if (bh)
 119                 BUFFER_TRACE(bh, "abort");
 120
 121         if (!handle->h_err)
 122                 handle->h_err = err;
 123
 124         if (is_handle_aborted(handle))
 125                 return;
 126
 127         printk(KERN_ERR "%s: aborting transaction: %s in %s\n",
 128                caller, errstr, err_fn);
 129
 130         journal_abort_handle(handle);
 131 }
 132
 133 /* Deal with the reporting of failure conditions on a filesystem such as
 134  * inconsistencies detected or read IO failures.
 135  *
 136  * On ext2, we can store the error state of the filesystem in the
 137  * superblock.  That is not possible on ext3, because we may have other
 138  * write ordering constraints on the superblock which prevent us from
 139  * writing it out straight away; and given that the journal is about to
 140  * be aborted, we can't rely on the current, or future, transactions to
 141  * write out the superblock safely.
 142  *
 143  * We'll just use the journal_abort() error code to record an error in
 144  * the journal instead.  On recovery, the journal will compain about
 145  * that error until we've noted it down and cleared it.
 146  */
 147
 148 static void ext3_handle_error(struct super_block *sb)
 149 {
 150         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 151
 152         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 153         es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
 154
 155         if (sb->s_flags & MS_RDONLY)
 156                 return;
 157
 158         if (test_opt (sb, ERRORS_RO)) {
 159                 printk (KERN_CRIT "Remounting filesystem read-only\n");
 160                 sb->s_flags |= MS_RDONLY;
 161         } else {
 162                 journal_t *journal = EXT3_SB(sb)->s_journal;
 163
 164                 EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 165                 if (journal)
 166                         journal_abort(journal, -EIO);
 167         }
 168         if (test_opt(sb, ERRORS_PANIC))
 169                 panic("EXT3-fs (device %s): panic forced after error\n",
 170                         sb->s_id);
 171         ext3_commit_super(sb, es, 1);
 172 }
 173
 174 void ext3_error (struct super_block * sb, const char * function,
 175                  const char * fmt, ...)
 176 {
 177         va_list args;
 178
 179         va_start(args, fmt);
 180         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 181         vprintk(fmt, args);
 182         printk("\n");
 183         va_end(args);
 184
 185         ext3_handle_error(sb);
 186 }
 187
 188 static const char *ext3_decode_error(struct super_block * sb, int errno,
 189                                      char nbuf[16])
 190 {
 191         char *errstr = NULL;
 192
 193         switch (errno) {
 194         case -EIO:
 195                 errstr = "IO failure";
 196                 break;
 197         case -ENOMEM:
 198                 errstr = "Out of memory";
 199                 break;
 200         case -EROFS:
 201                 if (!sb || EXT3_SB(sb)->s_journal->j_flags & JFS_ABORT)
 202                         errstr = "Journal has aborted";
 203                 else
 204                         errstr = "Readonly filesystem";
 205                 break;
 206         default:
 207                 /* If the caller passed in an extra buffer for unknown
 208                  * errors, textualise them now.  Else we just return
 209                  * NULL. */
 210                 if (nbuf) {
 211                         /* Check for truncated error codes... */
 212                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 213                                 errstr = nbuf;
 214                 }
 215                 break;
 216         }
 217
 218         return errstr;
 219 }
 220
 221 /* __ext3_std_error decodes expected errors from journaling functions
 222  * automatically and invokes the appropriate error response.  */
 223
 224 void __ext3_std_error (struct super_block * sb, const char * function,
 225                        int errno)
 226 {
 227         char nbuf[16];
 228         const char *errstr = ext3_decode_error(sb, errno, nbuf);
 229
 230         printk (KERN_CRIT "EXT3-fs error (device %s) in %s: %s\n",
 231                 sb->s_id, function, errstr);
 232
 233         ext3_handle_error(sb);
 234 }
 235
 236 /*
 237  * ext3_abort is a much stronger failure handler than ext3_error.  The
 238  * abort function may be used to deal with unrecoverable failures such
 239  * as journal IO errors or ENOMEM at a critical moment in log management.
 240  *
 241  * We unconditionally force the filesystem into an ABORT|READONLY state,
 242  * unless the error response on the fs has been set to panic in which
 243  * case we take the easy way out and panic immediately.
 244  */
 245
 246 void ext3_abort (struct super_block * sb, const char * function,
 247                  const char * fmt, ...)
 248 {
 249         va_list args;
 250
 251         printk (KERN_CRIT "ext3_abort called.\n");
 252
 253         va_start(args, fmt);
 254         printk(KERN_CRIT "EXT3-fs error (device %s): %s: ",sb->s_id, function);
 255         vprintk(fmt, args);
 256         printk("\n");
 257         va_end(args);
 258
 259         if (test_opt(sb, ERRORS_PANIC))
 260                 panic("EXT3-fs panic from previous error\n");
 261
 262         if (sb->s_flags & MS_RDONLY)
 263                 return;
 264
 265         printk(KERN_CRIT "Remounting filesystem read-only\n");
 266         EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
 267         sb->s_flags |= MS_RDONLY;
 268         EXT3_SB(sb)->s_mount_opt |= EXT3_MOUNT_ABORT;
 269         journal_abort(EXT3_SB(sb)->s_journal, -EIO);
 270 }
 271
 272 void ext3_warning (struct super_block * sb, const char * function,
 273                    const char * fmt, ...)
 274 {
 275         va_list args;
 276
 277         va_start(args, fmt);
 278         printk(KERN_WARNING "EXT3-fs warning (device %s): %s: ",
 279                sb->s_id, function);
 280         vprintk(fmt, args);
 281         printk("\n");
 282         va_end(args);
 283 }
 284
 285 void ext3_update_dynamic_rev(struct super_block *sb)
 286 {
 287         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
 288
 289         if (le32_to_cpu(es->s_rev_level) > EXT3_GOOD_OLD_REV)
 290                 return;
 291
 292         ext3_warning(sb, __FUNCTION__,
 293                      "updating to rev %d because of new feature flag, "
 294                      "running e2fsck is recommended",
 295                      EXT3_DYNAMIC_REV);
 296
 297         es->s_first_ino = cpu_to_le32(EXT3_GOOD_OLD_FIRST_INO);
 298         es->s_inode_size = cpu_to_le16(EXT3_GOOD_OLD_INODE_SIZE);
 299         es->s_rev_level = cpu_to_le32(EXT3_DYNAMIC_REV);
 300         /* leave es->s_feature_*compat flags alone */
 301         /* es->s_uuid will be set by e2fsck if empty */
 302
 303         /*
 304          * The rest of the superblock fields should be zero, and if not it
 305          * means they are likely already in use, so leave them alone.  We
 306          * can leave it up to e2fsck to clean up any inconsistencies there.
 307          */
 308 }
 309
 310 /*
 311  * Open the external journal device
 312  */
 313 static struct block_device *ext3_blkdev_get(dev_t dev)
 314 {
 315         struct block_device *bdev;
 316         char b[BDEVNAME_SIZE];
 317
 318         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
 319         if (IS_ERR(bdev))
 320                 goto fail;
 321         return bdev;
 322
 323 fail:
 324         printk(KERN_ERR "EXT3: failed to open journal device %s: %ld\n",
 325                         __bdevname(dev, b), PTR_ERR(bdev));
 326         return NULL;
 327 }
 328
 329 /*
 330  * Release the journal device
 331  */
 332 static int ext3_blkdev_put(struct block_device *bdev)
 333 {
 334         bd_release(bdev);
 335         return blkdev_put(bdev);
 336 }
 337
 338 static int ext3_blkdev_remove(struct ext3_sb_info *sbi)
 339 {
 340         struct block_device *bdev;
 341         int ret = -ENODEV;
 342
 343         bdev = sbi->journal_bdev;
 344         if (bdev) {
 345                 ret = ext3_blkdev_put(bdev);
 346                 sbi->journal_bdev = NULL;
 347         }
 348         return ret;
 349 }
 350
 351 static inline struct inode *orphan_list_entry(struct list_head *l)
 352 {
 353         return &list_entry(l, struct ext3_inode_info, i_orphan)->vfs_inode;
 354 }
 355
 356 static void dump_orphan_list(struct super_block *sb, struct ext3_sb_info *sbi)
 357 {
 358         struct list_head *l;
 359
 360         printk(KERN_ERR "sb orphan head is %d\n",
 361                le32_to_cpu(sbi->s_es->s_last_orphan));
 362
 363         printk(KERN_ERR "sb_info orphan list:\n");
 364         list_for_each(l, &sbi->s_orphan) {
 365                 struct inode *inode = orphan_list_entry(l);
 366                 printk(KERN_ERR "  "
 367                        "inode %s:%ld at %p: mode %o, nlink %d, next %d\n",
 368                        inode->i_sb->s_id, inode->i_ino, inode,
 369                        inode->i_mode, inode->i_nlink,
 370                        NEXT_ORPHAN(inode));
 371         }
 372 }
 373
 374 static void ext3_put_super (struct super_block * sb)
 375 {
 376         struct ext3_sb_info *sbi = EXT3_SB(sb);
 377         struct ext3_super_block *es = sbi->s_es;
 378         int i;
 379
 380         ext3_xattr_put_super(sb);
 381         journal_destroy(sbi->s_journal);
 382         if (!(sb->s_flags & MS_RDONLY)) {
 383                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
 384                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 385                 BUFFER_TRACE(sbi->s_sbh, "marking dirty");
 386                 mark_buffer_dirty(sbi->s_sbh);
 387                 ext3_commit_super(sb, es, 1);
 388         }
 389
 390         for (i = 0; i < sbi->s_gdb_count; i++)
 391                 brelse(sbi->s_group_desc[i]);
 392         kfree(sbi->s_group_desc);
 393         percpu_counter_destroy(&sbi->s_freeblocks_counter);
 394         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 395         percpu_counter_destroy(&sbi->s_dirs_counter);
 396         brelse(sbi->s_sbh);
 397 #ifdef CONFIG_QUOTA
 398         for (i = 0; i < MAXQUOTAS; i++) {
 399                 if (sbi->s_qf_names[i])
 400                         kfree(sbi->s_qf_names[i]);
 401         }
 402 #endif
 403
 404         /* Debugging code just in case the in-memory inode orphan list
 405          * isn't empty.  The on-disk one can be non-empty if we've
 406          * detected an error and taken the fs readonly, but the
 407          * in-memory list had better be clean by this point. */
 408         if (!list_empty(&sbi->s_orphan))
 409                 dump_orphan_list(sb, sbi);
 410         J_ASSERT(list_empty(&sbi->s_orphan));
 411
 412         invalidate_bdev(sb->s_bdev, 0);
 413         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 414                 /*
 415                  * Invalidate the journal device's buffers.  We don't want them
 416                  * floating about in memory - the physical journal device may
 417                  * hotswapped, and it breaks the `ro-after' testing code.
 418                  */
 419                 sync_blockdev(sbi->journal_bdev);
 420                 invalidate_bdev(sbi->journal_bdev, 0);
 421                 ext3_blkdev_remove(sbi);
 422         }
 423         sb->s_fs_info = NULL;
 424         kfree(sbi);
 425         return;
 426 }
 427
 428 static kmem_cache_t *ext3_inode_cachep;
 429
 430 /*
 431  * Called inside transaction, so use GFP_NOFS
 432  */
 433 static struct inode *ext3_alloc_inode(struct super_block *sb)
 434 {
 435         struct ext3_inode_info *ei;
 436
 437         ei = kmem_cache_alloc(ext3_inode_cachep, SLAB_NOFS);
 438         if (!ei)
 439                 return NULL;
 440 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 441         ei->i_acl = EXT3_ACL_NOT_CACHED;
 442         ei->i_default_acl = EXT3_ACL_NOT_CACHED;
 443 #endif
 444         ei->i_rsv_window.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
 445         ei->vfs_inode.i_version = 1;
 446         return &ei->vfs_inode;
 447 }
 448
 449 static void ext3_destroy_inode(struct inode *inode)
 450 {
 451         kmem_cache_free(ext3_inode_cachep, EXT3_I(inode));
 452 }
 453
 454 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
 455 {
 456         struct ext3_inode_info *ei = (struct ext3_inode_info *) foo;
 457
 458         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
 459             SLAB_CTOR_CONSTRUCTOR) {
 460                 INIT_LIST_HEAD(&ei->i_orphan);
 461 #ifdef CONFIG_EXT3_FS_XATTR
 462                 init_rwsem(&ei->xattr_sem);
 463 #endif
 464                 init_MUTEX(&ei->truncate_sem);
 465                 inode_init_once(&ei->vfs_inode);
 466         }
 467 }
 468
 469 static int init_inodecache(void)
 470 {
 471         ext3_inode_cachep = kmem_cache_create("ext3_inode_cache",
 472                                              sizeof(struct ext3_inode_info),
 473                                              0, SLAB_RECLAIM_ACCOUNT,
 474                                              init_once, NULL);
 475         if (ext3_inode_cachep == NULL)
 476                 return -ENOMEM;
 477         return 0;
 478 }
 479
 480 static void destroy_inodecache(void)
 481 {
 482         if (kmem_cache_destroy(ext3_inode_cachep))
 483                 printk(KERN_INFO "ext3_inode_cache: not all structures were freed\n");
 484 }
 485
 486 static void ext3_clear_inode(struct inode *inode)
 487 {
 488 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 489        if (EXT3_I(inode)->i_acl &&
 490            EXT3_I(inode)->i_acl != EXT3_ACL_NOT_CACHED) {
 491                posix_acl_release(EXT3_I(inode)->i_acl);
 492                EXT3_I(inode)->i_acl = EXT3_ACL_NOT_CACHED;
 493        }
 494        if (EXT3_I(inode)->i_default_acl &&
 495            EXT3_I(inode)->i_default_acl != EXT3_ACL_NOT_CACHED) {
 496                posix_acl_release(EXT3_I(inode)->i_default_acl);
 497                EXT3_I(inode)->i_default_acl = EXT3_ACL_NOT_CACHED;
 498        }
 499 #endif
 500         ext3_discard_reservation(inode);
 501 }
 502
 503 #ifdef CONFIG_QUOTA
 504
 505 #define QTYPE2NAME(t) ((t)==USRQUOTA?"user":"group")
 506 #define QTYPE2MOPT(on, t) ((t)==USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
 507
 508 static int ext3_dquot_initialize(struct inode *inode, int type);
 509 static int ext3_dquot_drop(struct inode *inode);
 510 static int ext3_write_dquot(struct dquot *dquot);
 511 static int ext3_acquire_dquot(struct dquot *dquot);
 512 static int ext3_release_dquot(struct dquot *dquot);
 513 static int ext3_mark_dquot_dirty(struct dquot *dquot);
 514 static int ext3_write_info(struct super_block *sb, int type);
 515 static int ext3_quota_on(struct super_block *sb, int type, int format_id, char *path);
 516 static int ext3_quota_on_mount(struct super_block *sb, int type);
 517 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
 518                                size_t len, loff_t off);
 519 static ssize_t ext3_quota_write(struct super_block *sb, int type,
 520                                 const char *data, size_t len, loff_t off);
 521
 522 static struct dquot_operations ext3_quota_operations = {
 523         .initialize     = ext3_dquot_initialize,
 524         .drop           = ext3_dquot_drop,
 525         .alloc_space    = dquot_alloc_space,
 526         .alloc_inode    = dquot_alloc_inode,
 527         .free_space     = dquot_free_space,
 528         .free_inode     = dquot_free_inode,
 529         .transfer       = dquot_transfer,
 530         .write_dquot    = ext3_write_dquot,
 531         .acquire_dquot  = ext3_acquire_dquot,
 532         .release_dquot  = ext3_release_dquot,
 533         .mark_dirty     = ext3_mark_dquot_dirty,
 534         .write_info     = ext3_write_info
 535 };
 536
 537 static struct quotactl_ops ext3_qctl_operations = {
 538         .quota_on       = ext3_quota_on,
 539         .quota_off      = vfs_quota_off,
 540         .quota_sync     = vfs_quota_sync,
 541         .get_info       = vfs_get_dqinfo,
 542         .set_info       = vfs_set_dqinfo,
 543         .get_dqblk      = vfs_get_dqblk,
 544         .set_dqblk      = vfs_set_dqblk
 545 };
 546 #endif
 547
 548 static struct super_operations ext3_sops = {
 549         .alloc_inode    = ext3_alloc_inode,
 550         .destroy_inode  = ext3_destroy_inode,
 551         .read_inode     = ext3_read_inode,
 552         .write_inode    = ext3_write_inode,
 553         .dirty_inode    = ext3_dirty_inode,
 554         .delete_inode   = ext3_delete_inode,
 555         .put_super      = ext3_put_super,
 556         .write_super    = ext3_write_super,
 557         .sync_fs        = ext3_sync_fs,
 558         .write_super_lockfs = ext3_write_super_lockfs,
 559         .unlockfs       = ext3_unlockfs,
 560         .statfs         = ext3_statfs,
 561         .remount_fs     = ext3_remount,
 562         .clear_inode    = ext3_clear_inode,
 563 #ifdef CONFIG_QUOTA
 564         .quota_read     = ext3_quota_read,
 565         .quota_write    = ext3_quota_write,
 566 #endif
 567 };
 568
 569 struct dentry *ext3_get_parent(struct dentry *child);
 570 static struct export_operations ext3_export_ops = {
 571         .get_parent = ext3_get_parent,
 572 };
 573
 574 enum {
 575         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
 576         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
 577         Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug, Opt_oldalloc, Opt_orlov,
 578         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
 579         Opt_reservation, Opt_noreservation, Opt_noload,
 580         Opt_commit, Opt_journal_update, Opt_journal_inum,
 581         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
 582         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
 583         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0,
 584         Opt_tagxid, Opt_barrier, Opt_ignore, Opt_err, Opt_resize,
 585 };
 586
 587 static match_table_t tokens = {
 588         {Opt_bsd_df, "bsddf"},
 589         {Opt_minix_df, "minixdf"},
 590         {Opt_grpid, "grpid"},
 591         {Opt_grpid, "bsdgroups"},
 592         {Opt_nogrpid, "nogrpid"},
 593         {Opt_nogrpid, "sysvgroups"},
 594         {Opt_resgid, "resgid=%u"},
 595         {Opt_resuid, "resuid=%u"},
 596         {Opt_sb, "sb=%u"},
 597         {Opt_err_cont, "errors=continue"},
 598         {Opt_err_panic, "errors=panic"},
 599         {Opt_err_ro, "errors=remount-ro"},
 600         {Opt_nouid32, "nouid32"},
 601         {Opt_nocheck, "nocheck"},
 602         {Opt_nocheck, "check=none"},
 603         {Opt_check, "check"},
 604         {Opt_debug, "debug"},
 605         {Opt_oldalloc, "oldalloc"},
 606         {Opt_orlov, "orlov"},
 607         {Opt_user_xattr, "user_xattr"},
 608         {Opt_nouser_xattr, "nouser_xattr"},
 609         {Opt_acl, "acl"},
 610         {Opt_noacl, "noacl"},
 611         {Opt_reservation, "reservation"},
 612         {Opt_noreservation, "noreservation"},
 613         {Opt_noload, "noload"},
 614         {Opt_commit, "commit=%u"},
 615         {Opt_journal_update, "journal=update"},
 616         {Opt_journal_inum, "journal=%u"},
 617         {Opt_abort, "abort"},
 618         {Opt_data_journal, "data=journal"},
 619         {Opt_data_ordered, "data=ordered"},
 620         {Opt_data_writeback, "data=writeback"},
 621         {Opt_offusrjquota, "usrjquota="},
 622         {Opt_usrjquota, "usrjquota=%s"},
 623         {Opt_offgrpjquota, "grpjquota="},
 624         {Opt_grpjquota, "grpjquota=%s"},
 625         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
 626         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
 627         {Opt_tagxid, "tagxid"},
 628         {Opt_ignore, "grpquota"},
 629         {Opt_ignore, "noquota"},
 630         {Opt_ignore, "quota"},
 631         {Opt_ignore, "usrquota"},
 632         {Opt_barrier, "barrier=%u"},
 633         {Opt_err, NULL},
 634         {Opt_resize, "resize"},
 635 };
 636
 637 static unsigned long get_sb_block(void **data)
 638 {
 639         unsigned long   sb_block;
 640         char            *options = (char *) *data;
 641
 642         if (!options || strncmp(options, "sb=", 3) != 0)
 643                 return 1;       /* Default location */
 644         options += 3;
 645         sb_block = simple_strtoul(options, &options, 0);
 646         if (*options && *options != ',') {
 647                 printk("EXT3-fs: Invalid sb specification: %s\n",
 648                        (char *) *data);
 649                 return 1;
 650         }
 651         if (*options == ',')
 652                 options++;
 653         *data = (void *) options;
 654         return sb_block;
 655 }
 656
 657 static int parse_options (char * options, struct super_block *sb,
 658                           unsigned long * inum, unsigned long *n_blocks_count, int is_remount)
 659 {
 660         struct ext3_sb_info *sbi = EXT3_SB(sb);
 661         char * p;
 662         substring_t args[MAX_OPT_ARGS];
 663         int data_opt = 0;
 664         int option;
 665 #ifdef CONFIG_QUOTA
 666         int qtype;
 667         char *qname;
 668 #endif
 669
 670         if (!options)
 671                 return 1;
 672
 673         while ((p = strsep (&options, ",")) != NULL) {
 674                 int token;
 675                 if (!*p)
 676                         continue;
 677
 678                 token = match_token(p, tokens, args);
 679                 switch (token) {
 680                 case Opt_bsd_df:
 681                         clear_opt (sbi->s_mount_opt, MINIX_DF);
 682                         break;
 683                 case Opt_minix_df:
 684                         set_opt (sbi->s_mount_opt, MINIX_DF);
 685                         break;
 686                 case Opt_grpid:
 687                         set_opt (sbi->s_mount_opt, GRPID);
 688                         break;
 689                 case Opt_nogrpid:
 690                         clear_opt (sbi->s_mount_opt, GRPID);
 691                         break;
 692                 case Opt_resuid:
 693                         if (match_int(&args[0], &option))
 694                                 return 0;
 695                         sbi->s_resuid = option;
 696                         break;
 697                 case Opt_resgid:
 698                         if (match_int(&args[0], &option))
 699                                 return 0;
 700                         sbi->s_resgid = option;
 701                         break;
 702                 case Opt_sb:
 703                         /* handled by get_sb_block() instead of here */
 704                         /* *sb_block = match_int(&args[0]); */
 705                         break;
 706                 case Opt_err_panic:
 707                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 708                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 709                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
 710                         break;
 711                 case Opt_err_ro:
 712                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
 713                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 714                         set_opt (sbi->s_mount_opt, ERRORS_RO);
 715                         break;
 716                 case Opt_err_cont:
 717                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
 718                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
 719                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
 720                         break;
 721                 case Opt_nouid32:
 722                         set_opt (sbi->s_mount_opt, NO_UID32);
 723                         break;
 724 #ifndef CONFIG_INOXID_NONE
 725                 case Opt_tagxid:
 726                         if (is_remount) {
 727                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 728                                        "tagxid on remount\n");
 729                                 return 0;
 730                         }
 731                         set_opt (sbi->s_mount_opt, TAG_XID);
 732                         break;
 733 #endif
 734                 case Opt_check:
 735 #ifdef CONFIG_EXT3_CHECK
 736                         set_opt (sbi->s_mount_opt, CHECK);
 737 #else
 738                         printk(KERN_ERR
 739                                "EXT3 Check option not supported\n");
 740 #endif
 741                         break;
 742                 case Opt_nocheck:
 743                         clear_opt (sbi->s_mount_opt, CHECK);
 744                         break;
 745                 case Opt_debug:
 746                         set_opt (sbi->s_mount_opt, DEBUG);
 747                         break;
 748                 case Opt_oldalloc:
 749                         set_opt (sbi->s_mount_opt, OLDALLOC);
 750                         break;
 751                 case Opt_orlov:
 752                         clear_opt (sbi->s_mount_opt, OLDALLOC);
 753                         break;
 754 #ifdef CONFIG_EXT3_FS_XATTR
 755                 case Opt_user_xattr:
 756                         set_opt (sbi->s_mount_opt, XATTR_USER);
 757                         break;
 758                 case Opt_nouser_xattr:
 759                         clear_opt (sbi->s_mount_opt, XATTR_USER);
 760                         break;
 761 #else
 762                 case Opt_user_xattr:
 763                 case Opt_nouser_xattr:
 764                         printk("EXT3 (no)user_xattr options not supported\n");
 765                         break;
 766 #endif
 767 #ifdef CONFIG_EXT3_FS_POSIX_ACL
 768                 case Opt_acl:
 769                         set_opt(sbi->s_mount_opt, POSIX_ACL);
 770                         break;
 771                 case Opt_noacl:
 772                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
 773                         break;
 774 #else
 775                 case Opt_acl:
 776                 case Opt_noacl:
 777                         printk("EXT3 (no)acl options not supported\n");
 778                         break;
 779 #endif
 780                 case Opt_reservation:
 781                         set_opt(sbi->s_mount_opt, RESERVATION);
 782                         break;
 783                 case Opt_noreservation:
 784                         clear_opt(sbi->s_mount_opt, RESERVATION);
 785                         break;
 786                 case Opt_journal_update:
 787                         /* @@@ FIXME */
 788                         /* Eventually we will want to be able to create
 789                            a journal file here.  For now, only allow the
 790                            user to specify an existing inode to be the
 791                            journal file. */
 792                         if (is_remount) {
 793                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 794                                        "journal on remount\n");
 795                                 return 0;
 796                         }
 797                         set_opt (sbi->s_mount_opt, UPDATE_JOURNAL);
 798                         break;
 799                 case Opt_journal_inum:
 800                         if (is_remount) {
 801                                 printk(KERN_ERR "EXT3-fs: cannot specify "
 802                                        "journal on remount\n");
 803                                 return 0;
 804                         }
 805                         if (match_int(&args[0], &option))
 806                                 return 0;
 807                         *inum = option;
 808                         break;
 809                 case Opt_noload:
 810                         set_opt (sbi->s_mount_opt, NOLOAD);
 811                         break;
 812                 case Opt_commit:
 813                         if (match_int(&args[0], &option))
 814                                 return 0;
 815                         if (option < 0)
 816                                 return 0;
 817                         if (option == 0)
 818                                 option = JBD_DEFAULT_MAX_COMMIT_AGE;
 819                         sbi->s_commit_interval = HZ * option;
 820                         break;
 821                 case Opt_data_journal:
 822                         data_opt = EXT3_MOUNT_JOURNAL_DATA;
 823                         goto datacheck;
 824                 case Opt_data_ordered:
 825                         data_opt = EXT3_MOUNT_ORDERED_DATA;
 826                         goto datacheck;
 827                 case Opt_data_writeback:
 828                         data_opt = EXT3_MOUNT_WRITEBACK_DATA;
 829                 datacheck:
 830                         if (is_remount) {
 831                                 if ((sbi->s_mount_opt & EXT3_MOUNT_DATA_FLAGS)
 832                                                 != data_opt) {
 833                                         printk(KERN_ERR
 834                                                 "EXT3-fs: cannot change data "
 835                                                 "mode on remount\n");
 836                                         return 0;
 837                                 }
 838                         } else {
 839                                 sbi->s_mount_opt &= ~EXT3_MOUNT_DATA_FLAGS;
 840                                 sbi->s_mount_opt |= data_opt;
 841                         }
 842                         break;
 843 #ifdef CONFIG_QUOTA
 844                 case Opt_usrjquota:
 845                         qtype = USRQUOTA;
 846                         goto set_qf_name;
 847                 case Opt_grpjquota:
 848                         qtype = GRPQUOTA;
 849 set_qf_name:
 850                         if (sb_any_quota_enabled(sb)) {
 851                                 printk(KERN_ERR
 852                                         "EXT3-fs: Cannot change journalled "
 853                                         "quota options when quota turned on.\n");
 854                                 return 0;
 855                         }
 856                         qname = match_strdup(&args[0]);
 857                         if (!qname) {
 858                                 printk(KERN_ERR
 859                                         "EXT3-fs: not enough memory for "
 860                                         "storing quotafile name.\n");
 861                                 return 0;
 862                         }
 863                         if (sbi->s_qf_names[qtype] &&
 864                             strcmp(sbi->s_qf_names[qtype], qname)) {
 865                                 printk(KERN_ERR
 866                                         "EXT3-fs: %s quota file already "
 867                                         "specified.\n", QTYPE2NAME(qtype));
 868                                 kfree(qname);
 869                                 return 0;
 870                         }
 871                         sbi->s_qf_names[qtype] = qname;
 872                         if (strchr(sbi->s_qf_names[qtype], '/')) {
 873                                 printk(KERN_ERR
 874                                         "EXT3-fs: quotafile must be on "
 875                                         "filesystem root.\n");
 876                                 kfree(sbi->s_qf_names[qtype]);
 877                                 sbi->s_qf_names[qtype] = NULL;
 878                                 return 0;
 879                         }
 880                         break;
 881                 case Opt_offusrjquota:
 882                         qtype = USRQUOTA;
 883                         goto clear_qf_name;
 884                 case Opt_offgrpjquota:
 885                         qtype = GRPQUOTA;
 886 clear_qf_name:
 887                         if (sb_any_quota_enabled(sb)) {
 888                                 printk(KERN_ERR "EXT3-fs: Cannot change "
 889                                         "journalled quota options when "
 890                                         "quota turned on.\n");
 891                                 return 0;
 892                         }
 893                         if (sbi->s_qf_names[qtype]) {
 894                                 kfree(sbi->s_qf_names[qtype]);
 895                                 sbi->s_qf_names[qtype] = NULL;
 896                         }
 897                         break;
 898                 case Opt_jqfmt_vfsold:
 899                         sbi->s_jquota_fmt = QFMT_VFS_OLD;
 900                         break;
 901                 case Opt_jqfmt_vfsv0:
 902                         sbi->s_jquota_fmt = QFMT_VFS_V0;
 903                         break;
 904 #else
 905                 case Opt_usrjquota:
 906                 case Opt_grpjquota:
 907                 case Opt_offusrjquota:
 908                 case Opt_offgrpjquota:
 909                 case Opt_jqfmt_vfsold:
 910                 case Opt_jqfmt_vfsv0:
 911                         printk(KERN_ERR
 912                                 "EXT3-fs: journalled quota options not "
 913                                 "supported.\n");
 914                         break;
 915 #endif
 916                 case Opt_abort:
 917                         set_opt(sbi->s_mount_opt, ABORT);
 918                         break;
 919                 case Opt_barrier:
 920                         if (match_int(&args[0], &option))
 921                                 return 0;
 922                         if (option)
 923                                 set_opt(sbi->s_mount_opt, BARRIER);
 924                         else
 925                                 clear_opt(sbi->s_mount_opt, BARRIER);
 926                         break;
 927                 case Opt_ignore:
 928                         break;
 929                 case Opt_resize:
 930                         if (!n_blocks_count) {
 931                                 printk("EXT3-fs: resize option only available "
 932                                         "for remount\n");
 933                                 return 0;
 934                         }
 935                         match_int(&args[0], &option);
 936                         *n_blocks_count = option;
 937                         break;
 938                 default:
 939                         printk (KERN_ERR
 940                                 "EXT3-fs: Unrecognized mount option \"%s\" "
 941                                 "or missing value\n", p);
 942                         return 0;
 943                 }
 944         }
 945 #ifdef CONFIG_QUOTA
 946         if (!sbi->s_jquota_fmt && (sbi->s_qf_names[USRQUOTA] ||
 947             sbi->s_qf_names[GRPQUOTA])) {
 948                 printk(KERN_ERR
 949                         "EXT3-fs: journalled quota format not specified.\n");
 950                 return 0;
 951         }
 952 #endif
 953
 954         return 1;
 955 }
 956
 957 static int ext3_setup_super(struct super_block *sb, struct ext3_super_block *es,
 958                             int read_only)
 959 {
 960         struct ext3_sb_info *sbi = EXT3_SB(sb);
 961         int res = 0;
 962
 963         if (le32_to_cpu(es->s_rev_level) > EXT3_MAX_SUPP_REV) {
 964                 printk (KERN_ERR "EXT3-fs warning: revision level too high, "
 965                         "forcing read-only mode\n");
 966                 res = MS_RDONLY;
 967         }
 968         if (read_only)
 969                 return res;
 970         if (!(sbi->s_mount_state & EXT3_VALID_FS))
 971                 printk (KERN_WARNING "EXT3-fs warning: mounting unchecked fs, "
 972                         "running e2fsck is recommended\n");
 973         else if ((sbi->s_mount_state & EXT3_ERROR_FS))
 974                 printk (KERN_WARNING
 975                         "EXT3-fs warning: mounting fs with errors, "
 976                         "running e2fsck is recommended\n");
 977         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
 978                  le16_to_cpu(es->s_mnt_count) >=
 979                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
 980                 printk (KERN_WARNING
 981                         "EXT3-fs warning: maximal mount count reached, "
 982                         "running e2fsck is recommended\n");
 983         else if (le32_to_cpu(es->s_checkinterval) &&
 984                 (le32_to_cpu(es->s_lastcheck) +
 985                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
 986                 printk (KERN_WARNING
 987                         "EXT3-fs warning: checktime reached, "
 988                         "running e2fsck is recommended\n");
 989 #if 0
 990                 /* @@@ We _will_ want to clear the valid bit if we find
 991                    inconsistencies, to force a fsck at reboot.  But for
 992                    a plain journaled filesystem we can keep it set as
 993                    valid forever! :) */
 994         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) & ~EXT3_VALID_FS);
 995 #endif
 996         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
 997                 es->s_max_mnt_count = cpu_to_le16(EXT3_DFL_MAX_MNT_COUNT);
 998         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
 999         es->s_mtime = cpu_to_le32(get_seconds());
1000         ext3_update_dynamic_rev(sb);
1001         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1002
1003         ext3_commit_super(sb, es, 1);
1004         if (test_opt(sb, DEBUG))
1005                 printk(KERN_INFO "[EXT3 FS bs=%lu, gc=%lu, "
1006                                 "bpg=%lu, ipg=%lu, mo=%04lx]\n",
1007                         sb->s_blocksize,
1008                         sbi->s_groups_count,
1009                         EXT3_BLOCKS_PER_GROUP(sb),
1010                         EXT3_INODES_PER_GROUP(sb),
1011                         sbi->s_mount_opt);
1012
1013         printk(KERN_INFO "EXT3 FS on %s, ", sb->s_id);
1014         if (EXT3_SB(sb)->s_journal->j_inode == NULL) {
1015                 char b[BDEVNAME_SIZE];
1016
1017                 printk("external journal on %s\n",
1018                         bdevname(EXT3_SB(sb)->s_journal->j_dev, b));
1019         } else {
1020                 printk("internal journal\n");
1021         }
1022 #ifdef CONFIG_EXT3_CHECK
1023         if (test_opt (sb, CHECK)) {
1024                 ext3_check_blocks_bitmap (sb);
1025                 ext3_check_inodes_bitmap (sb);
1026         }
1027 #endif
1028         return res;
1029 }
1030
1031 /* Called at mount-time, super-block is locked */
1032 static int ext3_check_descriptors (struct super_block * sb)
1033 {
1034         struct ext3_sb_info *sbi = EXT3_SB(sb);
1035         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
1036         struct ext3_group_desc * gdp = NULL;
1037         int desc_block = 0;
1038         int i;
1039
1040         ext3_debug ("Checking group descriptors");
1041
1042         for (i = 0; i < sbi->s_groups_count; i++)
1043         {
1044                 if ((i % EXT3_DESC_PER_BLOCK(sb)) == 0)
1045                         gdp = (struct ext3_group_desc *)
1046                                         sbi->s_group_desc[desc_block++]->b_data;
1047                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
1048                     le32_to_cpu(gdp->bg_block_bitmap) >=
1049                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1050                 {
1051                         ext3_error (sb, "ext3_check_descriptors",
1052                                     "Block bitmap for group %d"
1053                                     " not in group (block %lu)!",
1054                                     i, (unsigned long)
1055                                         le32_to_cpu(gdp->bg_block_bitmap));
1056                         return 0;
1057                 }
1058                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
1059                     le32_to_cpu(gdp->bg_inode_bitmap) >=
1060                                 block + EXT3_BLOCKS_PER_GROUP(sb))
1061                 {
1062                         ext3_error (sb, "ext3_check_descriptors",
1063                                     "Inode bitmap for group %d"
1064                                     " not in group (block %lu)!",
1065                                     i, (unsigned long)
1066                                         le32_to_cpu(gdp->bg_inode_bitmap));
1067                         return 0;
1068                 }
1069                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
1070                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
1071                     block + EXT3_BLOCKS_PER_GROUP(sb))
1072                 {
1073                         ext3_error (sb, "ext3_check_descriptors",
1074                                     "Inode table for group %d"
1075                                     " not in group (block %lu)!",
1076                                     i, (unsigned long)
1077                                         le32_to_cpu(gdp->bg_inode_table));
1078                         return 0;
1079                 }
1080                 block += EXT3_BLOCKS_PER_GROUP(sb);
1081                 gdp++;
1082         }
1083
1084         sbi->s_es->s_free_blocks_count=cpu_to_le32(ext3_count_free_blocks(sb));
1085         sbi->s_es->s_free_inodes_count=cpu_to_le32(ext3_count_free_inodes(sb));
1086         return 1;
1087 }
1088
1089
1090 /* ext3_orphan_cleanup() walks a singly-linked list of inodes (starting at
1091  * the superblock) which were deleted from all directories, but held open by
1092  * a process at the time of a crash.  We walk the list and try to delete these
1093  * inodes at recovery time (only with a read-write filesystem).
1094  *
1095  * In order to keep the orphan inode chain consistent during traversal (in
1096  * case of crash during recovery), we link each inode into the superblock
1097  * orphan list_head and handle it the same way as an inode deletion during
1098  * normal operation (which journals the operations for us).
1099  *
1100  * We only do an iget() and an iput() on each inode, which is very safe if we
1101  * accidentally point at an in-use or already deleted inode.  The worst that
1102  * can happen in this case is that we get a "bit already cleared" message from
1103  * ext3_free_inode().  The only reason we would point at a wrong inode is if
1104  * e2fsck was run on this filesystem, and it must have already done the orphan
1105  * inode cleanup for us, so we can safely abort without any further action.
1106  */
1107 static void ext3_orphan_cleanup (struct super_block * sb,
1108                                  struct ext3_super_block * es)
1109 {
1110         unsigned int s_flags = sb->s_flags;
1111         int nr_orphans = 0, nr_truncates = 0;
1112 #ifdef CONFIG_QUOTA
1113         int i;
1114 #endif
1115         if (!es->s_last_orphan) {
1116                 jbd_debug(4, "no orphan inodes to clean up\n");
1117                 return;
1118         }
1119
1120         if (EXT3_SB(sb)->s_mount_state & EXT3_ERROR_FS) {
1121                 if (es->s_last_orphan)
1122                         jbd_debug(1, "Errors on filesystem, "
1123                                   "clearing orphan list.\n");
1124                 es->s_last_orphan = 0;
1125                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
1126                 return;
1127         }
1128
1129         if (s_flags & MS_RDONLY) {
1130                 printk(KERN_INFO "EXT3-fs: %s: orphan cleanup on readonly fs\n",
1131                        sb->s_id);
1132                 sb->s_flags &= ~MS_RDONLY;
1133         }
1134 #ifdef CONFIG_QUOTA
1135         /* Needed for iput() to work correctly and not trash data */
1136         sb->s_flags |= MS_ACTIVE;
1137         /* Turn on quotas so that they are updated correctly */
1138         for (i = 0; i < MAXQUOTAS; i++) {
1139                 if (EXT3_SB(sb)->s_qf_names[i]) {
1140                         int ret = ext3_quota_on_mount(sb, i);
1141                         if (ret < 0)
1142                                 printk(KERN_ERR
1143                                         "EXT3-fs: Cannot turn on journalled "
1144                                         "quota: error %d\n", ret);
1145                 }
1146         }
1147 #endif
1148
1149         while (es->s_last_orphan) {
1150                 struct inode *inode;
1151
1152                 if (!(inode =
1153                       ext3_orphan_get(sb, le32_to_cpu(es->s_last_orphan)))) {
1154                         es->s_last_orphan = 0;
1155                         break;
1156                 }
1157
1158                 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1159                 DQUOT_INIT(inode);
1160                 if (inode->i_nlink) {
1161                         printk(KERN_DEBUG
1162                                 "%s: truncating inode %ld to %Ld bytes\n",
1163                                 __FUNCTION__, inode->i_ino, inode->i_size);
1164                         jbd_debug(2, "truncating inode %ld to %Ld bytes\n",
1165                                   inode->i_ino, inode->i_size);
1166                         ext3_truncate(inode);
1167                         nr_truncates++;
1168                 } else {
1169                         printk(KERN_DEBUG
1170                                 "%s: deleting unreferenced inode %ld\n",
1171                                 __FUNCTION__, inode->i_ino);
1172                         jbd_debug(2, "deleting unreferenced inode %ld\n",
1173                                   inode->i_ino);
1174                         nr_orphans++;
1175                 }
1176                 iput(inode);  /* The delete magic happens here! */
1177         }
1178
1179 #define PLURAL(x) (x), ((x)==1) ? "" : "s"
1180
1181         if (nr_orphans)
1182                 printk(KERN_INFO "EXT3-fs: %s: %d orphan inode%s deleted\n",
1183                        sb->s_id, PLURAL(nr_orphans));
1184         if (nr_truncates)
1185                 printk(KERN_INFO "EXT3-fs: %s: %d truncate%s cleaned up\n",
1186                        sb->s_id, PLURAL(nr_truncates));
1187 #ifdef CONFIG_QUOTA
1188         /* Turn quotas off */
1189         for (i = 0; i < MAXQUOTAS; i++) {
1190                 if (sb_dqopt(sb)->files[i])
1191                         vfs_quota_off(sb, i);
1192         }
1193 #endif
1194         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
1195 }
1196
1197 #define log2(n) ffz(~(n))
1198
1199 /*
1200  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
1201  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
1202  * We need to be 1 filesystem block less than the 2^32 sector limit.
1203  */
1204 static loff_t ext3_max_size(int bits)
1205 {
1206         loff_t res = EXT3_NDIR_BLOCKS;
1207         res += 1LL << (bits-2);
1208         res += 1LL << (2*(bits-2));
1209         res += 1LL << (3*(bits-2));
1210         res <<= bits;
1211         if (res > (512LL << 32) - (1 << bits))
1212                 res = (512LL << 32) - (1 << bits);
1213         return res;
1214 }
1215
1216 static unsigned long descriptor_loc(struct super_block *sb,
1217                                     unsigned long logic_sb_block,
1218                                     int nr)
1219 {
1220         struct ext3_sb_info *sbi = EXT3_SB(sb);
1221         unsigned long bg, first_data_block, first_meta_bg;
1222         int has_super = 0;
1223
1224         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1225         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
1226
1227         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_META_BG) ||
1228             nr < first_meta_bg)
1229                 return (logic_sb_block + nr + 1);
1230         bg = sbi->s_desc_per_block * nr;
1231         if (ext3_bg_has_super(sb, bg))
1232                 has_super = 1;
1233         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
1234 }
1235
1236
1237 static int ext3_fill_super (struct super_block *sb, void *data, int silent)
1238 {
1239         struct buffer_head * bh;
1240         struct ext3_super_block *es = NULL;
1241         struct ext3_sb_info *sbi;
1242         unsigned long block;
1243         unsigned long sb_block = get_sb_block(&data);
1244         unsigned long logic_sb_block;
1245         unsigned long offset = 0;
1246         unsigned long journal_inum = 0;
1247         unsigned long def_mount_opts;
1248         struct inode *root;
1249         int blocksize;
1250         int hblock;
1251         int db_count;
1252         int i;
1253         int needs_recovery;
1254         __le32 features;
1255
1256         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
1257         if (!sbi)
1258                 return -ENOMEM;
1259         sb->s_fs_info = sbi;
1260         memset(sbi, 0, sizeof(*sbi));
1261         sbi->s_mount_opt = 0;
1262         sbi->s_resuid = EXT3_DEF_RESUID;
1263         sbi->s_resgid = EXT3_DEF_RESGID;
1264
1265         unlock_kernel();
1266
1267         blocksize = sb_min_blocksize(sb, EXT3_MIN_BLOCK_SIZE);
1268         if (!blocksize) {
1269                 printk(KERN_ERR "EXT3-fs: unable to set blocksize\n");
1270                 goto out_fail;
1271         }
1272
1273         /*
1274          * The ext3 superblock will not be buffer aligned for other than 1kB
1275          * block sizes.  We need to calculate the offset from buffer start.
1276          */
1277         if (blocksize != EXT3_MIN_BLOCK_SIZE) {
1278                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1279                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1280         } else {
1281                 logic_sb_block = sb_block;
1282         }
1283
1284         if (!(bh = sb_bread(sb, logic_sb_block))) {
1285                 printk (KERN_ERR "EXT3-fs: unable to read superblock\n");
1286                 goto out_fail;
1287         }
1288         /*
1289          * Note: s_es must be initialized as soon as possible because
1290          *       some ext3 macro-instructions depend on its value
1291          */
1292         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1293         sbi->s_es = es;
1294         sb->s_magic = le16_to_cpu(es->s_magic);
1295         if (sb->s_magic != EXT3_SUPER_MAGIC)
1296                 goto cantfind_ext3;
1297
1298         /* Set defaults before we parse the mount options */
1299         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1300         if (def_mount_opts & EXT3_DEFM_DEBUG)
1301                 set_opt(sbi->s_mount_opt, DEBUG);
1302         if (def_mount_opts & EXT3_DEFM_BSDGROUPS)
1303                 set_opt(sbi->s_mount_opt, GRPID);
1304         if (def_mount_opts & EXT3_DEFM_UID16)
1305                 set_opt(sbi->s_mount_opt, NO_UID32);
1306         if (def_mount_opts & EXT3_DEFM_XATTR_USER)
1307                 set_opt(sbi->s_mount_opt, XATTR_USER);
1308         if (def_mount_opts & EXT3_DEFM_ACL)
1309                 set_opt(sbi->s_mount_opt, POSIX_ACL);
1310         if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_DATA)
1311                 sbi->s_mount_opt |= EXT3_MOUNT_JOURNAL_DATA;
1312         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_ORDERED)
1313                 sbi->s_mount_opt |= EXT3_MOUNT_ORDERED_DATA;
1314         else if ((def_mount_opts & EXT3_DEFM_JMODE) == EXT3_DEFM_JMODE_WBACK)
1315                 sbi->s_mount_opt |= EXT3_MOUNT_WRITEBACK_DATA;
1316
1317         if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_PANIC)
1318                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
1319         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT3_ERRORS_RO)
1320                 set_opt(sbi->s_mount_opt, ERRORS_RO);
1321
1322         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
1323         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
1324
1325         set_opt(sbi->s_mount_opt, RESERVATION);
1326
1327         if (!parse_options ((char *) data, sb, &journal_inum, NULL, 0))
1328                 goto failed_mount;
1329
1330         if (EXT3_SB(sb)->s_mount_opt & EXT3_MOUNT_TAG_XID)
1331                 sb->s_flags |= MS_TAGXID;
1332
1333         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1334                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1335
1336         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV &&
1337             (EXT3_HAS_COMPAT_FEATURE(sb, ~0U) ||
1338              EXT3_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
1339              EXT3_HAS_INCOMPAT_FEATURE(sb, ~0U)))
1340                 printk(KERN_WARNING
1341                        "EXT3-fs warning: feature flags set on rev 0 fs, "
1342                        "running e2fsck is recommended\n");
1343         /*
1344          * Check feature flags regardless of the revision level, since we
1345          * previously didn't change the revision level when setting the flags,
1346          * so there is a chance incompat flags are set on a rev 0 filesystem.
1347          */
1348         features = EXT3_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP);
1349         if (features) {
1350                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount because of "
1351                        "unsupported optional features (%x).\n",
1352                        sb->s_id, le32_to_cpu(features));
1353                 goto failed_mount;
1354         }
1355         features = EXT3_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP);
1356         if (!(sb->s_flags & MS_RDONLY) && features) {
1357                 printk(KERN_ERR "EXT3-fs: %s: couldn't mount RDWR because of "
1358                        "unsupported optional features (%x).\n",
1359                        sb->s_id, le32_to_cpu(features));
1360                 goto failed_mount;
1361         }
1362         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
1363
1364         if (blocksize < EXT3_MIN_BLOCK_SIZE ||
1365             blocksize > EXT3_MAX_BLOCK_SIZE) {
1366                 printk(KERN_ERR
1367                        "EXT3-fs: Unsupported filesystem blocksize %d on %s.\n",
1368                        blocksize, sb->s_id);
1369                 goto failed_mount;
1370         }
1371
1372         hblock = bdev_hardsect_size(sb->s_bdev);
1373         if (sb->s_blocksize != blocksize) {
1374                 /*
1375                  * Make sure the blocksize for the filesystem is larger
1376                  * than the hardware sectorsize for the machine.
1377                  */
1378                 if (blocksize < hblock) {
1379                         printk(KERN_ERR "EXT3-fs: blocksize %d too small for "
1380                                "device blocksize %d.\n", blocksize, hblock);
1381                         goto failed_mount;
1382                 }
1383
1384                 brelse (bh);
1385                 sb_set_blocksize(sb, blocksize);
1386                 logic_sb_block = (sb_block * EXT3_MIN_BLOCK_SIZE) / blocksize;
1387                 offset = (sb_block * EXT3_MIN_BLOCK_SIZE) % blocksize;
1388                 bh = sb_bread(sb, logic_sb_block);
1389                 if (!bh) {
1390                         printk(KERN_ERR
1391                                "EXT3-fs: Can't read superblock on 2nd try.\n");
1392                         goto failed_mount;
1393                 }
1394                 es = (struct ext3_super_block *)(((char *)bh->b_data) + offset);
1395                 sbi->s_es = es;
1396                 if (es->s_magic != cpu_to_le16(EXT3_SUPER_MAGIC)) {
1397                         printk (KERN_ERR
1398                                 "EXT3-fs: Magic mismatch, very weird !\n");
1399                         goto failed_mount;
1400                 }
1401         }
1402
1403         sb->s_maxbytes = ext3_max_size(sb->s_blocksize_bits);
1404
1405         if (le32_to_cpu(es->s_rev_level) == EXT3_GOOD_OLD_REV) {
1406                 sbi->s_inode_size = EXT3_GOOD_OLD_INODE_SIZE;
1407                 sbi->s_first_ino = EXT3_GOOD_OLD_FIRST_INO;
1408         } else {
1409                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
1410                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
1411                 if ((sbi->s_inode_size < EXT3_GOOD_OLD_INODE_SIZE) ||
1412                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
1413                     (sbi->s_inode_size > blocksize)) {
1414                         printk (KERN_ERR
1415                                 "EXT3-fs: unsupported inode size: %d\n",
1416                                 sbi->s_inode_size);
1417                         goto failed_mount;
1418                 }
1419         }
1420         sbi->s_frag_size = EXT3_MIN_FRAG_SIZE <<
1421                                    le32_to_cpu(es->s_log_frag_size);
1422         if (blocksize != sbi->s_frag_size) {
1423                 printk(KERN_ERR
1424                        "EXT3-fs: fragsize %lu != blocksize %u (unsupported)\n",
1425                        sbi->s_frag_size, blocksize);
1426                 goto failed_mount;
1427         }
1428         sbi->s_frags_per_block = 1;
1429         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
1430         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
1431         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
1432         if (EXT3_INODE_SIZE(sb) == 0)
1433                 goto cantfind_ext3;
1434         sbi->s_inodes_per_block = blocksize / EXT3_INODE_SIZE(sb);
1435         if (sbi->s_inodes_per_block == 0)
1436                 goto cantfind_ext3;
1437         sbi->s_itb_per_group = sbi->s_inodes_per_group /
1438                                         sbi->s_inodes_per_block;
1439         sbi->s_desc_per_block = blocksize / sizeof(struct ext3_group_desc);
1440         sbi->s_sbh = bh;
1441         sbi->s_mount_state = le16_to_cpu(es->s_state);
1442         sbi->s_addr_per_block_bits = log2(EXT3_ADDR_PER_BLOCK(sb));
1443         sbi->s_desc_per_block_bits = log2(EXT3_DESC_PER_BLOCK(sb));
1444         for (i=0; i < 4; i++)
1445                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
1446         sbi->s_def_hash_version = es->s_def_hash_version;
1447
1448         if (sbi->s_blocks_per_group > blocksize * 8) {
1449                 printk (KERN_ERR
1450                         "EXT3-fs: #blocks per group too big: %lu\n",
1451                         sbi->s_blocks_per_group);
1452                 goto failed_mount;
1453         }
1454         if (sbi->s_frags_per_group > blocksize * 8) {
1455                 printk (KERN_ERR
1456                         "EXT3-fs: #fragments per group too big: %lu\n",
1457                         sbi->s_frags_per_group);
1458                 goto failed_mount;
1459         }
1460         if (sbi->s_inodes_per_group > blocksize * 8) {
1461                 printk (KERN_ERR
1462                         "EXT3-fs: #inodes per group too big: %lu\n",
1463                         sbi->s_inodes_per_group);
1464                 goto failed_mount;
1465         }
1466
1467         if (EXT3_BLOCKS_PER_GROUP(sb) == 0)
1468                 goto cantfind_ext3;
1469         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
1470                                le32_to_cpu(es->s_first_data_block) +
1471                                EXT3_BLOCKS_PER_GROUP(sb) - 1) /
1472                               EXT3_BLOCKS_PER_GROUP(sb);
1473         db_count = (sbi->s_groups_count + EXT3_DESC_PER_BLOCK(sb) - 1) /
1474                    EXT3_DESC_PER_BLOCK(sb);
1475         sbi->s_group_desc = kmalloc(db_count * sizeof (struct buffer_head *),
1476                                     GFP_KERNEL);
1477         if (sbi->s_group_desc == NULL) {
1478                 printk (KERN_ERR "EXT3-fs: not enough memory\n");
1479                 goto failed_mount;
1480         }
1481
1482         percpu_counter_init(&sbi->s_freeblocks_counter);
1483         percpu_counter_init(&sbi->s_freeinodes_counter);
1484         percpu_counter_init(&sbi->s_dirs_counter);
1485         bgl_lock_init(&sbi->s_blockgroup_lock);
1486
1487         for (i = 0; i < db_count; i++) {
1488                 block = descriptor_loc(sb, logic_sb_block, i);
1489                 sbi->s_group_desc[i] = sb_bread(sb, block);
1490                 if (!sbi->s_group_desc[i]) {
1491                         printk (KERN_ERR "EXT3-fs: "
1492                                 "can't read group descriptor %d\n", i);
1493                         db_count = i;
1494                         goto failed_mount2;
1495                 }
1496         }
1497         if (!ext3_check_descriptors (sb)) {
1498                 printk (KERN_ERR "EXT3-fs: group descriptors corrupted !\n");
1499                 goto failed_mount2;
1500         }
1501         sbi->s_gdb_count = db_count;
1502         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
1503         spin_lock_init(&sbi->s_next_gen_lock);
1504         /* per fileystem reservation list head & lock */
1505         spin_lock_init(&sbi->s_rsv_window_lock);
1506         sbi->s_rsv_window_root = RB_ROOT;
1507         /* Add a single, static dummy reservation to the start of the
1508          * reservation window list --- it gives us a placeholder for
1509          * append-at-start-of-list which makes the allocation logic
1510          * _much_ simpler. */
1511         sbi->s_rsv_window_head.rsv_start = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1512         sbi->s_rsv_window_head.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
1513         atomic_set(&sbi->s_rsv_window_head.rsv_alloc_hit, 0);
1514         atomic_set(&sbi->s_rsv_window_head.rsv_goal_size, 0);
1515         ext3_rsv_window_add(sb, &sbi->s_rsv_window_head);
1516
1517         /*
1518          * set up enough so that it can read an inode
1519          */
1520         sb->s_op = &ext3_sops;
1521         sb->s_export_op = &ext3_export_ops;
1522         sb->s_xattr = ext3_xattr_handlers;
1523 #ifdef CONFIG_QUOTA
1524         sb->s_qcop = &ext3_qctl_operations;
1525         sb->dq_op = &ext3_quota_operations;
1526 #endif
1527         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
1528
1529         sb->s_root = NULL;
1530
1531         needs_recovery = (es->s_last_orphan != 0 ||
1532                           EXT3_HAS_INCOMPAT_FEATURE(sb,
1533                                     EXT3_FEATURE_INCOMPAT_RECOVER));
1534
1535         /*
1536          * The first inode we look at is the journal inode.  Don't try
1537          * root first: it may be modified in the journal!
1538          */
1539         if (!test_opt(sb, NOLOAD) &&
1540             EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1541                 if (ext3_load_journal(sb, es))
1542                         goto failed_mount2;
1543         } else if (journal_inum) {
1544                 if (ext3_create_journal(sb, es, journal_inum))
1545                         goto failed_mount2;
1546         } else {
1547                 if (!silent)
1548                         printk (KERN_ERR
1549                                 "ext3: No journal on filesystem on %s\n",
1550                                 sb->s_id);
1551                 goto failed_mount2;
1552         }
1553
1554         /* We have now updated the journal if required, so we can
1555          * validate the data journaling mode. */
1556         switch (test_opt(sb, DATA_FLAGS)) {
1557         case 0:
1558                 /* No mode set, assume a default based on the journal
1559                    capabilities: ORDERED_DATA if the journal can
1560                    cope, else JOURNAL_DATA */
1561                 if (journal_check_available_features
1562                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE))
1563                         set_opt(sbi->s_mount_opt, ORDERED_DATA);
1564                 else
1565                         set_opt(sbi->s_mount_opt, JOURNAL_DATA);
1566                 break;
1567
1568         case EXT3_MOUNT_ORDERED_DATA:
1569         case EXT3_MOUNT_WRITEBACK_DATA:
1570                 if (!journal_check_available_features
1571                     (sbi->s_journal, 0, 0, JFS_FEATURE_INCOMPAT_REVOKE)) {
1572                         printk(KERN_ERR "EXT3-fs: Journal does not support "
1573                                "requested data journaling mode\n");
1574                         goto failed_mount3;
1575                 }
1576         default:
1577                 break;
1578         }
1579
1580         /*
1581          * The journal_load will have done any necessary log recovery,
1582          * so we can safely mount the rest of the filesystem now.
1583          */
1584
1585         root = iget(sb, EXT3_ROOT_INO);
1586         sb->s_root = d_alloc_root(root);
1587         if (!sb->s_root) {
1588                 printk(KERN_ERR "EXT3-fs: get root inode failed\n");
1589                 iput(root);
1590                 goto failed_mount3;
1591         }
1592         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
1593                 dput(sb->s_root);
1594                 sb->s_root = NULL;
1595                 printk(KERN_ERR "EXT3-fs: corrupt root inode, run e2fsck\n");
1596                 goto failed_mount3;
1597         }
1598
1599         ext3_setup_super (sb, es, sb->s_flags & MS_RDONLY);
1600         /*
1601          * akpm: core read_super() calls in here with the superblock locked.
1602          * That deadlocks, because orphan cleanup needs to lock the superblock
1603          * in numerous places.  Here we just pop the lock - it's relatively
1604          * harmless, because we are now ready to accept write_super() requests,
1605          * and aviro says that's the only reason for hanging onto the
1606          * superblock lock.
1607          */
1608         EXT3_SB(sb)->s_mount_state |= EXT3_ORPHAN_FS;
1609         ext3_orphan_cleanup(sb, es);
1610         EXT3_SB(sb)->s_mount_state &= ~EXT3_ORPHAN_FS;
1611         if (needs_recovery)
1612                 printk (KERN_INFO "EXT3-fs: recovery complete.\n");
1613         ext3_mark_recovery_complete(sb, es);
1614         printk (KERN_INFO "EXT3-fs: mounted filesystem with %s data mode.\n",
1615                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ? "journal":
1616                 test_opt(sb,DATA_FLAGS) == EXT3_MOUNT_ORDERED_DATA ? "ordered":
1617                 "writeback");
1618
1619         percpu_counter_mod(&sbi->s_freeblocks_counter,
1620                 ext3_count_free_blocks(sb));
1621         percpu_counter_mod(&sbi->s_freeinodes_counter,
1622                 ext3_count_free_inodes(sb));
1623         percpu_counter_mod(&sbi->s_dirs_counter,
1624                 ext3_count_dirs(sb));
1625
1626         lock_kernel();
1627         return 0;
1628
1629 cantfind_ext3:
1630         if (!silent)
1631                 printk(KERN_ERR "VFS: Can't find ext3 filesystem on dev %s.\n",
1632                        sb->s_id);
1633         goto failed_mount;
1634
1635 failed_mount3:
1636         journal_destroy(sbi->s_journal);
1637 failed_mount2:
1638         for (i = 0; i < db_count; i++)
1639                 brelse(sbi->s_group_desc[i]);
1640         kfree(sbi->s_group_desc);
1641 failed_mount:
1642 #ifdef CONFIG_QUOTA
1643         for (i = 0; i < MAXQUOTAS; i++)
1644                 kfree(sbi->s_qf_names[i]);
1645 #endif
1646         ext3_blkdev_remove(sbi);
1647         brelse(bh);
1648 out_fail:
1649         sb->s_fs_info = NULL;
1650         kfree(sbi);
1651         lock_kernel();
1652         return -EINVAL;
1653 }
1654
1655 /*
1656  * Setup any per-fs journal parameters now.  We'll do this both on
1657  * initial mount, once the journal has been initialised but before we've
1658  * done any recovery; and again on any subsequent remount.
1659  */
1660 static void ext3_init_journal_params(struct super_block *sb, journal_t *journal)
1661 {
1662         struct ext3_sb_info *sbi = EXT3_SB(sb);
1663
1664         if (sbi->s_commit_interval)
1665                 journal->j_commit_interval = sbi->s_commit_interval;
1666         /* We could also set up an ext3-specific default for the commit
1667          * interval here, but for now we'll just fall back to the jbd
1668          * default. */
1669
1670         spin_lock(&journal->j_state_lock);
1671         if (test_opt(sb, BARRIER))
1672                 journal->j_flags |= JFS_BARRIER;
1673         else
1674                 journal->j_flags &= ~JFS_BARRIER;
1675         spin_unlock(&journal->j_state_lock);
1676 }
1677
1678 static journal_t *ext3_get_journal(struct super_block *sb, int journal_inum)
1679 {
1680         struct inode *journal_inode;
1681         journal_t *journal;
1682
1683         /* First, test for the existence of a valid inode on disk.  Bad
1684          * things happen if we iget() an unused inode, as the subsequent
1685          * iput() will try to delete it. */
1686
1687         journal_inode = iget(sb, journal_inum);
1688         if (!journal_inode) {
1689                 printk(KERN_ERR "EXT3-fs: no journal found.\n");
1690                 return NULL;
1691         }
1692         if (!journal_inode->i_nlink) {
1693                 make_bad_inode(journal_inode);
1694                 iput(journal_inode);
1695                 printk(KERN_ERR "EXT3-fs: journal inode is deleted.\n");
1696                 return NULL;
1697         }
1698
1699         jbd_debug(2, "Journal inode found at %p: %Ld bytes\n",
1700                   journal_inode, journal_inode->i_size);
1701         if (is_bad_inode(journal_inode) || !S_ISREG(journal_inode->i_mode)) {
1702                 printk(KERN_ERR "EXT3-fs: invalid journal inode.\n");
1703                 iput(journal_inode);
1704                 return NULL;
1705         }
1706
1707         journal = journal_init_inode(journal_inode);
1708         if (!journal) {
1709                 printk(KERN_ERR "EXT3-fs: Could not load journal inode\n");
1710                 iput(journal_inode);
1711                 return NULL;
1712         }
1713         journal->j_private = sb;
1714         ext3_init_journal_params(sb, journal);
1715         return journal;
1716 }
1717
1718 static journal_t *ext3_get_dev_journal(struct super_block *sb,
1719                                        dev_t j_dev)
1720 {
1721         struct buffer_head * bh;
1722         journal_t *journal;
1723         int start;
1724         int len;
1725         int hblock, blocksize;
1726         unsigned long sb_block;
1727         unsigned long offset;
1728         struct ext3_super_block * es;
1729         struct block_device *bdev;
1730
1731         bdev = ext3_blkdev_get(j_dev);
1732         if (bdev == NULL)
1733                 return NULL;
1734
1735         if (bd_claim(bdev, sb)) {
1736                 printk(KERN_ERR
1737                         "EXT3: failed to claim external journal device.\n");
1738                 blkdev_put(bdev);
1739                 return NULL;
1740         }
1741
1742         blocksize = sb->s_blocksize;
1743         hblock = bdev_hardsect_size(bdev);
1744         if (blocksize < hblock) {
1745                 printk(KERN_ERR
1746                         "EXT3-fs: blocksize too small for journal device.\n");
1747                 goto out_bdev;
1748         }
1749
1750         sb_block = EXT3_MIN_BLOCK_SIZE / blocksize;
1751         offset = EXT3_MIN_BLOCK_SIZE % blocksize;
1752         set_blocksize(bdev, blocksize);
1753         if (!(bh = __bread(bdev, sb_block, blocksize))) {
1754                 printk(KERN_ERR "EXT3-fs: couldn't read superblock of "
1755                        "external journal\n");
1756                 goto out_bdev;
1757         }
1758
1759         es = (struct ext3_super_block *) (((char *)bh->b_data) + offset);
1760         if ((le16_to_cpu(es->s_magic) != EXT3_SUPER_MAGIC) ||
1761             !(le32_to_cpu(es->s_feature_incompat) &
1762               EXT3_FEATURE_INCOMPAT_JOURNAL_DEV)) {
1763                 printk(KERN_ERR "EXT3-fs: external journal has "
1764                                         "bad superblock\n");
1765                 brelse(bh);
1766                 goto out_bdev;
1767         }
1768
1769         if (memcmp(EXT3_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
1770                 printk(KERN_ERR "EXT3-fs: journal UUID does not match\n");
1771                 brelse(bh);
1772                 goto out_bdev;
1773         }
1774
1775         len = le32_to_cpu(es->s_blocks_count);
1776         start = sb_block + 1;
1777         brelse(bh);     /* we're done with the superblock */
1778
1779         journal = journal_init_dev(bdev, sb->s_bdev,
1780                                         start, len, blocksize);
1781         if (!journal) {
1782                 printk(KERN_ERR "EXT3-fs: failed to create device journal\n");
1783                 goto out_bdev;
1784         }
1785         journal->j_private = sb;
1786         ll_rw_block(READ, 1, &journal->j_sb_buffer);
1787         wait_on_buffer(journal->j_sb_buffer);
1788         if (!buffer_uptodate(journal->j_sb_buffer)) {
1789                 printk(KERN_ERR "EXT3-fs: I/O error on journal device\n");
1790                 goto out_journal;
1791         }
1792         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
1793                 printk(KERN_ERR "EXT3-fs: External journal has more than one "
1794                                         "user (unsupported) - %d\n",
1795                         be32_to_cpu(journal->j_superblock->s_nr_users));
1796                 goto out_journal;
1797         }
1798         EXT3_SB(sb)->journal_bdev = bdev;
1799         ext3_init_journal_params(sb, journal);
1800         return journal;
1801 out_journal:
1802         journal_destroy(journal);
1803 out_bdev:
1804         ext3_blkdev_put(bdev);
1805         return NULL;
1806 }
1807
1808 static int ext3_load_journal(struct super_block * sb,
1809                              struct ext3_super_block * es)
1810 {
1811         journal_t *journal;
1812         int journal_inum = le32_to_cpu(es->s_journal_inum);
1813         dev_t journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
1814         int err = 0;
1815         int really_read_only;
1816
1817         really_read_only = bdev_read_only(sb->s_bdev);
1818
1819         /*
1820          * Are we loading a blank journal or performing recovery after a
1821          * crash?  For recovery, we need to check in advance whether we
1822          * can get read-write access to the device.
1823          */
1824
1825         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER)) {
1826                 if (sb->s_flags & MS_RDONLY) {
1827                         printk(KERN_INFO "EXT3-fs: INFO: recovery "
1828                                         "required on readonly filesystem.\n");
1829                         if (really_read_only) {
1830                                 printk(KERN_ERR "EXT3-fs: write access "
1831                                         "unavailable, cannot proceed.\n");
1832                                 return -EROFS;
1833                         }
1834                         printk (KERN_INFO "EXT3-fs: write access will "
1835                                         "be enabled during recovery.\n");
1836                 }
1837         }
1838
1839         if (journal_inum && journal_dev) {
1840                 printk(KERN_ERR "EXT3-fs: filesystem has both journal "
1841                        "and inode journals!\n");
1842                 return -EINVAL;
1843         }
1844
1845         if (journal_inum) {
1846                 if (!(journal = ext3_get_journal(sb, journal_inum)))
1847                         return -EINVAL;
1848         } else {
1849                 if (!(journal = ext3_get_dev_journal(sb, journal_dev)))
1850                         return -EINVAL;
1851         }
1852
1853         if (!really_read_only && test_opt(sb, UPDATE_JOURNAL)) {
1854                 err = journal_update_format(journal);
1855                 if (err)  {
1856                         printk(KERN_ERR "EXT3-fs: error updating journal.\n");
1857                         journal_destroy(journal);
1858                         return err;
1859                 }
1860         }
1861
1862         if (!EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER))
1863                 err = journal_wipe(journal, !really_read_only);
1864         if (!err)
1865                 err = journal_load(journal);
1866
1867         if (err) {
1868                 printk(KERN_ERR "EXT3-fs: error loading journal.\n");
1869                 journal_destroy(journal);
1870                 return err;
1871         }
1872
1873         EXT3_SB(sb)->s_journal = journal;
1874         ext3_clear_journal_err(sb, es);
1875         return 0;
1876 }
1877
1878 static int ext3_create_journal(struct super_block * sb,
1879                                struct ext3_super_block * es,
1880                                int journal_inum)
1881 {
1882         journal_t *journal;
1883
1884         if (sb->s_flags & MS_RDONLY) {
1885                 printk(KERN_ERR "EXT3-fs: readonly filesystem when trying to "
1886                                 "create journal.\n");
1887                 return -EROFS;
1888         }
1889
1890         if (!(journal = ext3_get_journal(sb, journal_inum)))
1891                 return -EINVAL;
1892
1893         printk(KERN_INFO "EXT3-fs: creating new journal on inode %d\n",
1894                journal_inum);
1895
1896         if (journal_create(journal)) {
1897                 printk(KERN_ERR "EXT3-fs: error creating journal.\n");
1898                 journal_destroy(journal);
1899                 return -EIO;
1900         }
1901
1902         EXT3_SB(sb)->s_journal = journal;
1903
1904         ext3_update_dynamic_rev(sb);
1905         EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1906         EXT3_SET_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL);
1907
1908         es->s_journal_inum = cpu_to_le32(journal_inum);
1909         sb->s_dirt = 1;
1910
1911         /* Make sure we flush the recovery flag to disk. */
1912         ext3_commit_super(sb, es, 1);
1913
1914         return 0;
1915 }
1916
1917 static void ext3_commit_super (struct super_block * sb,
1918                                struct ext3_super_block * es,
1919                                int sync)
1920 {
1921         struct buffer_head *sbh = EXT3_SB(sb)->s_sbh;
1922
1923         if (!sbh)
1924                 return;
1925         es->s_wtime = cpu_to_le32(get_seconds());
1926         es->s_free_blocks_count = cpu_to_le32(ext3_count_free_blocks(sb));
1927         es->s_free_inodes_count = cpu_to_le32(ext3_count_free_inodes(sb));
1928         BUFFER_TRACE(sbh, "marking dirty");
1929         mark_buffer_dirty(sbh);
1930         if (sync)
1931                 sync_dirty_buffer(sbh);
1932 }
1933
1934
1935 /*
1936  * Have we just finished recovery?  If so, and if we are mounting (or
1937  * remounting) the filesystem readonly, then we will end up with a
1938  * consistent fs on disk.  Record that fact.
1939  */
1940 static void ext3_mark_recovery_complete(struct super_block * sb,
1941                                         struct ext3_super_block * es)
1942 {
1943         journal_t *journal = EXT3_SB(sb)->s_journal;
1944
1945         journal_lock_updates(journal);
1946         journal_flush(journal);
1947         if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER) &&
1948             sb->s_flags & MS_RDONLY) {
1949                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
1950                 sb->s_dirt = 0;
1951                 ext3_commit_super(sb, es, 1);
1952         }
1953         journal_unlock_updates(journal);
1954 }
1955
1956 /*
1957  * If we are mounting (or read-write remounting) a filesystem whose journal
1958  * has recorded an error from a previous lifetime, move that error to the
1959  * main filesystem now.
1960  */
1961 static void ext3_clear_journal_err(struct super_block * sb,
1962                                    struct ext3_super_block * es)
1963 {
1964         journal_t *journal;
1965         int j_errno;
1966         const char *errstr;
1967
1968         journal = EXT3_SB(sb)->s_journal;
1969
1970         /*
1971          * Now check for any error status which may have been recorded in the
1972          * journal by a prior ext3_error() or ext3_abort()
1973          */
1974
1975         j_errno = journal_errno(journal);
1976         if (j_errno) {
1977                 char nbuf[16];
1978
1979                 errstr = ext3_decode_error(sb, j_errno, nbuf);
1980                 ext3_warning(sb, __FUNCTION__, "Filesystem error recorded "
1981                              "from previous mount: %s", errstr);
1982                 ext3_warning(sb, __FUNCTION__, "Marking fs in need of "
1983                              "filesystem check.");
1984
1985                 EXT3_SB(sb)->s_mount_state |= EXT3_ERROR_FS;
1986                 es->s_state |= cpu_to_le16(EXT3_ERROR_FS);
1987                 ext3_commit_super (sb, es, 1);
1988
1989                 journal_clear_err(journal);
1990         }
1991 }
1992
1993 /*
1994  * Force the running and committing transactions to commit,
1995  * and wait on the commit.
1996  */
1997 int ext3_force_commit(struct super_block *sb)
1998 {
1999         journal_t *journal;
2000         int ret;
2001
2002         if (sb->s_flags & MS_RDONLY)
2003                 return 0;
2004
2005         journal = EXT3_SB(sb)->s_journal;
2006         sb->s_dirt = 0;
2007         ret = ext3_journal_force_commit(journal);
2008         return ret;
2009 }
2010
2011 /*
2012  * Ext3 always journals updates to the superblock itself, so we don't
2013  * have to propagate any other updates to the superblock on disk at this
2014  * point.  Just start an async writeback to get the buffers on their way
2015  * to the disk.
2016  *
2017  * This implicitly triggers the writebehind on sync().
2018  */
2019
2020 static void ext3_write_super (struct super_block * sb)
2021 {
2022         if (down_trylock(&sb->s_lock) == 0)
2023                 BUG();
2024         sb->s_dirt = 0;
2025 }
2026
2027 static int ext3_sync_fs(struct super_block *sb, int wait)
2028 {
2029         tid_t target;
2030
2031         sb->s_dirt = 0;
2032         if (journal_start_commit(EXT3_SB(sb)->s_journal, &target)) {
2033                 if (wait)
2034                         log_wait_commit(EXT3_SB(sb)->s_journal, target);
2035         }
2036         return 0;
2037 }
2038
2039 /*
2040  * LVM calls this function before a (read-only) snapshot is created.  This
2041  * gives us a chance to flush the journal completely and mark the fs clean.
2042  */
2043 static void ext3_write_super_lockfs(struct super_block *sb)
2044 {
2045         sb->s_dirt = 0;
2046
2047         if (!(sb->s_flags & MS_RDONLY)) {
2048                 journal_t *journal = EXT3_SB(sb)->s_journal;
2049
2050                 /* Now we set up the journal barrier. */
2051                 journal_lock_updates(journal);
2052                 journal_flush(journal);
2053
2054                 /* Journal blocked and flushed, clear needs_recovery flag. */
2055                 EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2056                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2057         }
2058 }
2059
2060 /*
2061  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
2062  * flag here, even though the filesystem is not technically dirty yet.
2063  */
2064 static void ext3_unlockfs(struct super_block *sb)
2065 {
2066         if (!(sb->s_flags & MS_RDONLY)) {
2067                 lock_super(sb);
2068                 /* Reser the needs_recovery flag before the fs is unlocked. */
2069                 EXT3_SET_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
2070                 ext3_commit_super(sb, EXT3_SB(sb)->s_es, 1);
2071                 unlock_super(sb);
2072                 journal_unlock_updates(EXT3_SB(sb)->s_journal);
2073         }
2074 }
2075
2076 static int ext3_remount (struct super_block * sb, int * flags, char * data)
2077 {
2078         struct ext3_super_block * es;
2079         struct ext3_sb_info *sbi = EXT3_SB(sb);
2080         unsigned long tmp;
2081         unsigned long n_blocks_count = 0;
2082
2083         /*
2084          * Allow the "check" option to be passed as a remount option.
2085          */
2086         if (!parse_options(data, sb, &tmp, &n_blocks_count, 1))
2087                 return -EINVAL;
2088
2089         if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2090                 ext3_abort(sb, __FUNCTION__, "Abort forced by user");
2091
2092         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
2093                 ((sbi->s_mount_opt & EXT3_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
2094
2095         es = sbi->s_es;
2096
2097         ext3_init_journal_params(sb, sbi->s_journal);
2098
2099         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY) ||
2100                 n_blocks_count > le32_to_cpu(es->s_blocks_count)) {
2101                 if (sbi->s_mount_opt & EXT3_MOUNT_ABORT)
2102                         return -EROFS;
2103
2104                 if (*flags & MS_RDONLY) {
2105                         /*
2106                          * First of all, the unconditional stuff we have to do
2107                          * to disable replay of the journal when we next remount
2108                          */
2109                         sb->s_flags |= MS_RDONLY;
2110
2111                         /*
2112                          * OK, test if we are remounting a valid rw partition
2113                          * readonly, and if so set the rdonly flag and then
2114                          * mark the partition as valid again.
2115                          */
2116                         if (!(es->s_state & cpu_to_le16(EXT3_VALID_FS)) &&
2117                             (sbi->s_mount_state & EXT3_VALID_FS))
2118                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
2119
2120                         ext3_mark_recovery_complete(sb, es);
2121                 } else {
2122                         __le32 ret;
2123                         if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
2124                                         ~EXT3_FEATURE_RO_COMPAT_SUPP))) {
2125                                 printk(KERN_WARNING "EXT3-fs: %s: couldn't "
2126                                        "remount RDWR because of unsupported "
2127                                        "optional features (%x).\n",
2128                                        sb->s_id, le32_to_cpu(ret));
2129                                 return -EROFS;
2130                         }
2131                         /*
2132                          * Mounting a RDONLY partition read-write, so reread
2133                          * and store the current valid flag.  (It may have
2134                          * been changed by e2fsck since we originally mounted
2135                          * the partition.)
2136                          */
2137                         ext3_clear_journal_err(sb, es);
2138                         sbi->s_mount_state = le16_to_cpu(es->s_state);
2139                         if ((ret = ext3_group_extend(sb, es, n_blocks_count)))
2140                                 return ret;
2141                         if (!ext3_setup_super (sb, es, 0))
2142                                 sb->s_flags &= ~MS_RDONLY;
2143                 }
2144         }
2145         return 0;
2146 }
2147
2148 static int ext3_statfs (struct super_block * sb, struct kstatfs * buf)
2149 {
2150         struct ext3_super_block *es = EXT3_SB(sb)->s_es;
2151         unsigned long overhead;
2152         int i;
2153
2154         if (test_opt (sb, MINIX_DF))
2155                 overhead = 0;
2156         else {
2157                 unsigned long ngroups;
2158                 ngroups = EXT3_SB(sb)->s_groups_count;
2159                 smp_rmb();
2160
2161                 /*
2162                  * Compute the overhead (FS structures)
2163                  */
2164
2165                 /*
2166                  * All of the blocks before first_data_block are
2167                  * overhead
2168                  */
2169                 overhead = le32_to_cpu(es->s_first_data_block);
2170
2171                 /*
2172                  * Add the overhead attributed to the superblock and
2173                  * block group descriptors.  If the sparse superblocks
2174                  * feature is turned on, then not all groups have this.
2175                  */
2176                 for (i = 0; i < ngroups; i++) {
2177                         overhead += ext3_bg_has_super(sb, i) +
2178                                 ext3_bg_num_gdb(sb, i);
2179                         cond_resched();
2180                 }
2181
2182                 /*
2183                  * Every block group has an inode bitmap, a block
2184                  * bitmap, and an inode table.
2185                  */
2186                 overhead += (ngroups * (2 + EXT3_SB(sb)->s_itb_per_group));
2187         }
2188
2189         buf->f_type = EXT3_SUPER_MAGIC;
2190         buf->f_bsize = sb->s_blocksize;
2191         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
2192         buf->f_bfree = ext3_count_free_blocks (sb);
2193         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
2194         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
2195                 buf->f_bavail = 0;
2196         buf->f_files = le32_to_cpu(es->s_inodes_count);
2197         buf->f_ffree = ext3_count_free_inodes (sb);
2198         buf->f_namelen = EXT3_NAME_LEN;
2199         return 0;
2200 }
2201
2202 /* Helper function for writing quotas on sync - we need to start transaction before quota file
2203  * is locked for write. Otherwise the are possible deadlocks:
2204  * Process 1                         Process 2
2205  * ext3_create()                     quota_sync()
2206  *   journal_start()                   write_dquot()
2207  *   DQUOT_INIT()                        down(dqio_sem)
2208  *     down(dqio_sem)                    journal_start()
2209  *
2210  */
2211
2212 #ifdef CONFIG_QUOTA
2213
2214 static inline struct inode *dquot_to_inode(struct dquot *dquot)
2215 {
2216         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
2217 }
2218
2219 static int ext3_dquot_initialize(struct inode *inode, int type)
2220 {
2221         handle_t *handle;
2222         int ret, err;
2223
2224         /* We may create quota structure so we need to reserve enough blocks */
2225         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2226         if (IS_ERR(handle))
2227                 return PTR_ERR(handle);
2228         ret = dquot_initialize(inode, type);
2229         err = ext3_journal_stop(handle);
2230         if (!ret)
2231                 ret = err;
2232         return ret;
2233 }
2234
2235 static int ext3_dquot_drop(struct inode *inode)
2236 {
2237         handle_t *handle;
2238         int ret, err;
2239
2240         /* We may delete quota structure so we need to reserve enough blocks */
2241         handle = ext3_journal_start(inode, 2*EXT3_QUOTA_INIT_BLOCKS);
2242         if (IS_ERR(handle))
2243                 return PTR_ERR(handle);
2244         ret = dquot_drop(inode);
2245         err = ext3_journal_stop(handle);
2246         if (!ret)
2247                 ret = err;
2248         return ret;
2249 }
2250
2251 static int ext3_write_dquot(struct dquot *dquot)
2252 {
2253         int ret, err;
2254         handle_t *handle;
2255         struct inode *inode;
2256
2257         inode = dquot_to_inode(dquot);
2258         handle = ext3_journal_start(inode,
2259                                         EXT3_QUOTA_TRANS_BLOCKS);
2260         if (IS_ERR(handle))
2261                 return PTR_ERR(handle);
2262         ret = dquot_commit(dquot);
2263         err = ext3_journal_stop(handle);
2264         if (!ret)
2265                 ret = err;
2266         return ret;
2267 }
2268
2269 static int ext3_acquire_dquot(struct dquot *dquot)
2270 {
2271         int ret, err;
2272         handle_t *handle;
2273
2274         handle = ext3_journal_start(dquot_to_inode(dquot),
2275                                         EXT3_QUOTA_INIT_BLOCKS);
2276         if (IS_ERR(handle))
2277                 return PTR_ERR(handle);
2278         ret = dquot_acquire(dquot);
2279         err = ext3_journal_stop(handle);
2280         if (!ret)
2281                 ret = err;
2282         return ret;
2283 }
2284
2285 static int ext3_release_dquot(struct dquot *dquot)
2286 {
2287         int ret, err;
2288         handle_t *handle;
2289
2290         handle = ext3_journal_start(dquot_to_inode(dquot),
2291                                         EXT3_QUOTA_INIT_BLOCKS);
2292         if (IS_ERR(handle))
2293                 return PTR_ERR(handle);
2294         ret = dquot_release(dquot);
2295         err = ext3_journal_stop(handle);
2296         if (!ret)
2297                 ret = err;
2298         return ret;
2299 }
2300
2301 static int ext3_mark_dquot_dirty(struct dquot *dquot)
2302 {
2303         /* Are we journalling quotas? */
2304         if (EXT3_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
2305             EXT3_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
2306                 dquot_mark_dquot_dirty(dquot);
2307                 return ext3_write_dquot(dquot);
2308         } else {
2309                 return dquot_mark_dquot_dirty(dquot);
2310         }
2311 }
2312
2313 static int ext3_write_info(struct super_block *sb, int type)
2314 {
2315         int ret, err;
2316         handle_t *handle;
2317
2318         /* Data block + inode block */
2319         handle = ext3_journal_start(sb->s_root->d_inode, 2);
2320         if (IS_ERR(handle))
2321                 return PTR_ERR(handle);
2322         ret = dquot_commit_info(sb, type);
2323         err = ext3_journal_stop(handle);
2324         if (!ret)
2325                 ret = err;
2326         return ret;
2327 }
2328
2329 /*
2330  * Turn on quotas during mount time - we need to find
2331  * the quota file and such...
2332  */
2333 static int ext3_quota_on_mount(struct super_block *sb, int type)
2334 {
2335         int err;
2336         struct dentry *dentry;
2337         struct qstr name = { .name = EXT3_SB(sb)->s_qf_names[type],
2338                              .hash = 0,
2339                              .len = strlen(EXT3_SB(sb)->s_qf_names[type])};
2340
2341         dentry = lookup_hash(&name, sb->s_root);
2342         if (IS_ERR(dentry))
2343                 return PTR_ERR(dentry);
2344         err = vfs_quota_on_mount(type, EXT3_SB(sb)->s_jquota_fmt, dentry);
2345         /* Now invalidate and put the dentry - quota got its own reference
2346          * to inode and dentry has at least wrong hash so we had better
2347          * throw it away */
2348         d_invalidate(dentry);
2349         dput(dentry);
2350         return err;
2351 }
2352
2353 /*
2354  * Standard function to be called on quota_on
2355  */
2356 static int ext3_quota_on(struct super_block *sb, int type, int format_id,
2357                          char *path)
2358 {
2359         int err;
2360         struct nameidata nd;
2361
2362         /* Not journalling quota? */
2363         if (!EXT3_SB(sb)->s_qf_names[USRQUOTA] &&
2364             !EXT3_SB(sb)->s_qf_names[GRPQUOTA])
2365                 return vfs_quota_on(sb, type, format_id, path);
2366         err = path_lookup(path, LOOKUP_FOLLOW, &nd);
2367         if (err)
2368                 return err;
2369         /* Quotafile not on the same filesystem? */
2370         if (nd.mnt->mnt_sb != sb) {
2371                 path_release(&nd);
2372                 return -EXDEV;
2373         }
2374         /* Quotafile not of fs root? */
2375         if (nd.dentry->d_parent->d_inode != sb->s_root->d_inode)
2376                 printk(KERN_WARNING
2377                         "EXT3-fs: Quota file not on filesystem root. "
2378                         "Journalled quota will not work.\n");
2379         path_release(&nd);
2380         return vfs_quota_on(sb, type, format_id, path);
2381 }
2382
2383 /* Read data from quotafile - avoid pagecache and such because we cannot afford
2384  * acquiring the locks... As quota files are never truncated and quota code
2385  * itself serializes the operations (and noone else should touch the files)
2386  * we don't have to be afraid of races */
2387 static ssize_t ext3_quota_read(struct super_block *sb, int type, char *data,
2388                                size_t len, loff_t off)
2389 {
2390         struct inode *inode = sb_dqopt(sb)->files[type];
2391         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2392         int err = 0;
2393         int offset = off & (sb->s_blocksize - 1);
2394         int tocopy;
2395         size_t toread;
2396         struct buffer_head *bh;
2397         loff_t i_size = i_size_read(inode);
2398
2399         if (off > i_size)
2400                 return 0;
2401         if (off+len > i_size)
2402                 len = i_size-off;
2403         toread = len;
2404         while (toread > 0) {
2405                 tocopy = sb->s_blocksize - offset < toread ?
2406                                 sb->s_blocksize - offset : toread;
2407                 bh = ext3_bread(NULL, inode, blk, 0, &err);
2408                 if (err)
2409                         return err;
2410                 if (!bh)        /* A hole? */
2411                         memset(data, 0, tocopy);
2412                 else
2413                         memcpy(data, bh->b_data+offset, tocopy);
2414                 brelse(bh);
2415                 offset = 0;
2416                 toread -= tocopy;
2417                 data += tocopy;
2418                 blk++;
2419         }
2420         return len;
2421 }
2422
2423 /* Write to quotafile (we know the transaction is already started and has
2424  * enough credits) */
2425 static ssize_t ext3_quota_write(struct super_block *sb, int type,
2426                                 const char *data, size_t len, loff_t off)
2427 {
2428         struct inode *inode = sb_dqopt(sb)->files[type];
2429         sector_t blk = off >> EXT3_BLOCK_SIZE_BITS(sb);
2430         int err = 0;
2431         int offset = off & (sb->s_blocksize - 1);
2432         int tocopy;
2433         int journal_quota = EXT3_SB(sb)->s_qf_names[type] != NULL;
2434         size_t towrite = len;
2435         struct buffer_head *bh;
2436         handle_t *handle = journal_current_handle();
2437
2438         down(&inode->i_sem);
2439         while (towrite > 0) {
2440                 tocopy = sb->s_blocksize - offset < towrite ?
2441                                 sb->s_blocksize - offset : towrite;
2442                 bh = ext3_bread(handle, inode, blk, 1, &err);
2443                 if (!bh)
2444                         goto out;
2445                 if (journal_quota) {
2446                         err = ext3_journal_get_write_access(handle, bh);
2447                         if (err) {
2448                                 brelse(bh);
2449                                 goto out;
2450                         }
2451                 }
2452                 lock_buffer(bh);
2453                 memcpy(bh->b_data+offset, data, tocopy);
2454                 flush_dcache_page(bh->b_page);
2455                 unlock_buffer(bh);
2456                 if (journal_quota)
2457                         err = ext3_journal_dirty_metadata(handle, bh);
2458                 else {
2459                         /* Always do at least ordered writes for quotas */
2460                         err = ext3_journal_dirty_data(handle, bh);
2461                         mark_buffer_dirty(bh);
2462                 }
2463                 brelse(bh);
2464                 if (err)
2465                         goto out;
2466                 offset = 0;
2467                 towrite -= tocopy;
2468                 data += tocopy;
2469                 blk++;
2470         }
2471 out:
2472         if (len == towrite)
2473                 return err;
2474         if (inode->i_size < off+len-towrite) {
2475                 i_size_write(inode, off+len-towrite);
2476                 EXT3_I(inode)->i_disksize = inode->i_size;
2477         }
2478         inode->i_version++;
2479         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2480         ext3_mark_inode_dirty(handle, inode);
2481         up(&inode->i_sem);
2482         return len - towrite;
2483 }
2484
2485 #endif
2486
2487 static struct super_block *ext3_get_sb(struct file_system_type *fs_type,
2488         int flags, const char *dev_name, void *data)
2489 {
2490         return get_sb_bdev(fs_type, flags, dev_name, data, ext3_fill_super);
2491 }
2492
2493 static struct file_system_type ext3_fs_type = {
2494         .owner          = THIS_MODULE,
2495         .name           = "ext3",
2496         .get_sb         = ext3_get_sb,
2497         .kill_sb        = kill_block_super,
2498         .fs_flags       = FS_REQUIRES_DEV,
2499 };
2500
2501 static int __init init_ext3_fs(void)
2502 {
2503         int err = init_ext3_xattr();
2504         if (err)
2505                 return err;
2506         err = init_inodecache();
2507         if (err)
2508                 goto out1;
2509         err = register_filesystem(&ext3_fs_type);
2510         if (err)
2511                 goto out;
2512         return 0;
2513 out:
2514         destroy_inodecache();
2515 out1:
2516         exit_ext3_xattr();
2517         return err;
2518 }
2519
2520 static void __exit exit_ext3_fs(void)
2521 {
2522         unregister_filesystem(&ext3_fs_type);
2523         destroy_inodecache();
2524         exit_ext3_xattr();
2525 }
2526
2527 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
2528 MODULE_DESCRIPTION("Second Extended Filesystem with journaling extensions");
2529 MODULE_LICENSE("GPL");
2530 module_init(init_ext3_fs)
2531 module_exit(exit_ext3_fs)